Knowledge Discovery Pipeline for The Classification of Kinase Inhibitors.

Abstract

The human kinome, comprised of the kinase complement of the genome, constitutes approximately 2% of the entire genome, encoding a total of 538 distinct proteins. These kinases play a pivotal role in cellular regulation by catalyzing the addition of phosphate groups to diverse substrates, thereby intricately modulating the activation and deactivation of crucial regulatory and signaling pathways. Perturbations such as mutations, overexpression, and loss of function within kinases have been associated in a spectrum of diseases, prominently encompassing cancers, neurodegenerative disorders, metabolic and immune-related ailments. It is noteworthy that a significant proportion of approximately one third of therapeutic targets within the pharmaceutical domain are developed around kinases. An emerging paradigm that holds promise in this pursuit is computational drug discovery, augmented by the application of machine learning protocols. This synergistic approach has effectively modernized the drug discovery process, substantially curtailing the temporal demands of this intricate endeavor. Within the purview of this study, a comprehensive classification of kinases was undertaken, discerning them into two distinct categories: those with discernible links to disease pathogenesis, notably implicated in cancer and neurodegenerative disorders, and those with roles that do not inherently contribute to disease progression. A data matrix containing 9200 sequential, topological, and proto chemometric descriptors of 501 protein kinases was implied to train a set of classifiers resulting in Random Forest as the best classifier achieving 65% accuracy. The subsequent focus was directed towards screening protein kinase inhibitors to ascertain their potential to impede kinases associated with neurodegenerative maladies. Artificial neural network came out as the bast classifier (attaining 76% accuracy) out of an array of classifiers trained on the dataset of Morgan fingerprints of 490 inhibitors. The compounds that emerged from this screening exhibited promising inhibitory capabilities, positioning them as prime candidates for therapeutic deployment in the amelioration of neurodegenerative conditions.